Neurocognitive modeling meets deep learning: Understanding differences in cognition across the human life span (PI: Mischa von Krause; 2025-2028)
In an aging society, understanding age differences in cognition is becoming increasingly important. Mathematical process models of cognition offer an interesting approach to studying cognitive parameters across the human lifespan. By utilizing both response times and accuracy data obtained from cognitive tasks, models such as the diffusion decision model enable researchers to better understand what drives age differences in cognition, such as the well-established fact that older adults tend to show slower response times than younger adults. Perhaps rather surprisingly, model-driven analyses have often shown that these age differences in response times can mostly be explained by differences in speed-accuracy settings and the speed of non-decisional processes (encoding, motor response execution). Contrarily, the studies often found few age differences in the speed of evidence accumulation or mental speed. However, as can be expected for a relatively new field of studies, there are still open questions. In the proposed research programme, three important gaps in the literature will be addressed, thus helping us better understand when, how, and why age differences in cognitive parameters occur. First, the generalizability of findings across researcher degrees of freedom and demographics is still widely understudied. To address this gap, it is planned to conduct a multiverse analysis, comparing different model architectures, estimation procedures, and data cleaning methods; it is also planned to conduct a big data analysis comparing age effects across a wide range of demographics. Second, previous studies have mostly ignored possible dynamics in parameters within experimental blocks. In this research proposal, we plan to develop and apply a novel model of age differences in within-block dynamics based on very recent advances in deep learning based dynamic modeling. Third, findings on how age differences in cognitive parameters are linked to neurophysiological measures are unclear and sometimes puzzling. To address this gap, we plan to conduct both a meta-analysis quantitative summarizing previous findings, and to develop a new joint model of neurophysiology and cognition. Said model will then by validated and applied to a large dataset of twelve cognitive tasks. In sum, the planned endeavors across the three work packages should greatly enhance our knowledge of age differences in neurocognitive parameters: By expanding our knowledge on the generalizability of previous findings, by scrutinizing within-block dynamics, and by better understanding the neurophysiological basis of patterns in cognitive parameters across age groups. All three of the gaps will be addressed by utilizing state-of-the-art deep learning methods. These methods have only recently become available and, by greatly enlarging the scope of neurocognitive models that can be developed and applied, promise an important step forward in understand age differences in cognition.
GRK 2277: Statistical Modeling in Psychology (SMiP, 2017-2027)
The Research Training Group “Statistical Modeling in Psychology” (SMiP) aims to overcome the growing gap between statistical innovations and their use in psychological research. Through novel applications of complex statistical models and techniques in different content areas of psychology, and through the development of new methods of statistical modeling, parameter estimation and inference, the Research Training Group intends to raise the quality of research in psychology and to foster the transfer of statistical modeling approaches across subdisciplines of psychology. The SMiP research program covers a series of projects that combine substantive research questions from key areas of psychological inquiry with advanced techniques of statistical modeling. Substantive research questions include tests of new predictions in experimental research on cognition and motivation, in longitudinal research on affect dynamics and personality development, and in research on individual differences in cognitive abilities and personality traits. The research goals are pursued with state-of-the-art and newly developed statistical models and techniques that allow researchers to measure latent psychological states or processes, to test novel predictions, and to compare competing theoretical accounts in a coherent analytical framework. The individual projects are mutually related through overarching thematic research clusters that focus on modeling heterogeneity of psychological effects and processes, and on integrative and comprehensive tests of psychological hypotheses. The research program is interlinked with a qualification program that equips a new generation of researchers with extended theoretical knowledge and practical skills in using and developing advanced statistical methods for a professional career in academia or for a research-related position outside academia. The structured qualification program consists of core courses and elective courses, coaching units, and a joint supervision scheme. The SMiP research and qualification agenda is integrated in an international network, including regular courses given by international experts, visits of SMiP doctoral researchers at international research labs, and joint workshops and summer schools in collaboration with international doctoral training programs.
Lévy-Flight Models of Decision-Making (2020-2025)
In the last decades, the diffusion model (Ratcliff, 1978) became one the most popular models in cognitive psychology. The model proposes a mathematical account of cognitive processes underlying fast binary decisions. The present project challenges one of the core assumptions of the diffusion model, that is, the postulation that information accumulation follows a Wiener diffusion process. This process is a composition of a constant drift rate (i.e., a constant rate of evidence accumulation over time) and normal distributed noise. In the new Lévy-Flight account (Voss, Lerche, Mertens, & Voss, in press), the Gaussian noise distribution is replaced by a heavy-tailed distribution (e.g., a Cauchy distribution). This fundamentally changes the proposed evidence accumulation mechanism and the model's predictions. In contrast to the diffusion model, the new model assumes that evidence accumulation incorporates "jumps", that is, sudden large changes in the amount of accumulated evidence. This not only alters the shape of predicted response-time distributions, but also provides an explanation for the fact that in simple perceptual tasks errors are typically faster than correct responses (Luce, 1984). This project aims at (a) developing an efficient and user-friendly program for Lévy-Flight data analyses; (b) comparing results from diffusion model analyses and Lévy-Flight analyses and (c) applying the model to assess inter-individual differences in rigid vs. flexible decision-making styles. The project will help to improve understanding of processes underlying human decision-making.
Selbstregulation nach Frustration des Leistungsmotivs (2018)
gefördert durch das "Field of Focus 4: Self-Regulation and Regulation" der Universität Heidelberg
Die Motivtheorie (u.a. Brunstein, Schultheiss, & Grässman, 1998; Schüler, Sheldon, & Fröhlich, 2010) geht davon aus, dass Menschen mit einer hohen Motivausprägung extremer auf eine Befriedigung bzw. Frustration des jeweiligen Motives reagieren. Demnach sollten Menschen mit einer hohen Ausprägung des Leistungsmotivs positiver auf positives Leistungsfeedback reagieren und negativer auf negatives Feedback. In einigen korrelativen Studien konnte Unterstützung für diese These gefunden werden. In zwei Studien mit einem experimentellen Design und mit großen Stichprobengrößen (N = 150 bzw. N = 148) fand unsere Arbeitsgruppe jedoch ein hypothesenkonträres Muster: Nicht die hoch, sondern die niedrig Leistungsmotivierten reagierten auf negatives Leistungsfeedback - sowohl bei interindividueller (Studie 1) als auch bei intraindividueller Feedbackgabe (Studie 2) - besonders negativ.
In diesem FoF4-Projekt möchten wir der Ursache dieses unerwarteten Befundes nachgehen. Wir erwarten die Ursache dabei in Unterschieden in der Selbstregulation zwischen niedrig und hoch Leistungsmotivierten. Wir nehmen an, dass hoch Leistungsmotivierte zwar auf negatives Leistungsfeedback anfänglich negativer reagieren als niedrig Leistungsmotivierte, dass sie zugleich jedoch auch schneller effektive Strategien der Gegenregulation (Coping) einsetzen. Unterstützung für diese Gegenregulationshypothese liefern etwa Befunde, dass sich hoch Leistungsmotivierte besser an leistungsrelevante Erlebnisse erinnern (z.B. Woike, McLeod, & Goggin, 2003). Denkbar ist folglich, dass die hoch Leistungsmotivierten sich durch die Erinnerung an positive Leistungserlebnisse von dem negativen Feedback ablenken können und somit ihren anfänglichen negativen Affekt schnell herunter regulieren. Dies gelingt niedrig Leistungsmotivierten in geringerem Ausmaß.
Thema unseres Projekts ist die Untersuchung interindividueller Unterschiede in der Selbstregulation (kognitiv und emotional) nach negativem Leistungsfeedback. Hierbei werden wir verschiedene methodische Ansätze (Eyetracking, Mixed-Methods) verfolgen.
Modeling slow decisions: Validation of the Diffusion Model for a new type of tasks (2012-2018)
This project investigates the possibility to use stochastic diffusion models for the analysis of reaction time data from tasks with response latencies of several seconds. Diffusion models allow disentangling cognitive processes from binary decision tasks by mapping these on distinct parameters. The model provides, among other, valid measures of the rate of information accumulation, the individual response threshold, and the duration of encoding and response execution. Thus, this form of data analysis allows a detailed understanding of cognitive decision-making processes. Until now, fast decisions (of several hundred milliseconds), were considered a prerequisite for diffusion model analyses. However, such fast classification tasks have a low ecological validity for everyday decisions. We argue that the diffusion model is also suitable for tasks with longer phases of information accumulation or processing. Since the cognitive processes in such slower tasks are more typical for decision-making in everyday life, we expect a better criterion validity of the model parameters. In this project, first the validity of the diffusion model is tested for various experimental paradigms with moderate to high response latencies (Work Package 1). Then we investigate whether the drift parameter of the model, which reflects the rate of information accumulation, can be used as a diagnostic measure of intelligence (Work Package 2), and whether the threshold parameter can be used as a measure of impulsive decision-making behavior (Work Package 3).
Assessing stress in daily life: Ambulatory assessement of physiological indicators in an intensive longitudinal design (2016-2017)
gefördert durch die FRONTIER-Initiative der Universität Heidelberg
Deficient self-regulation in ADHD: Using the diffusion model to measure impulsive decision making (2014-2015)
gefördert durch das "Field of Focus 4: Self-Regulation and Regulation" der Universität Heidelberg
Regulation, Donation Behavior and Moral Perception: An Experimental Investigation (2014-2015)
Prerequisits for data analysis with stochastic diffusion models: Comparison of optimization criteria (2012-2015)
Mit stochastischen Diffusionsmodellen (Ratcliff, 1978) können kognitive Prozesse erfasst werden, die bei schnellen binären Entscheidungen ablaufen. Dabei werden die Reaktionszeitverteilungen von korrekten Antworten und Fehlern berücksichtigt um Parameter zu schätzen, die spezifische kognitive Prozesse (z.B. Geschwindigkeit der Informationsaufnahme; Menge der für eine Entscheidung berücksichtigten Information) abbilden. Durch diese Form der Analyse ist es möglich, spezifische Hypothesen über kognitive Prozesse bei der Bearbeitung einfacher Entscheidungsaufgabe zu testen. Im aktuellen Projekt wird die Effizienz und Robustheit von Schätzverfahren, die auf unterschiedlichen Optimierungskriterien beruhen (Chi-Square, Maximum-Likelihood, Kolmogorov-Smirnov) bei kleinen, mittleren und großen Datensätzen systematisch verglichen. Insbesondere soll untersucht werden, unter welchen Bedingungen Diffusionsmodellanalysen auch bei kleinen Datensätzen zu reliablen Ergebnissen kommen. Es sollen konkrete Empfehlungen abgeleitet werden, welche Datensätze für Diffusionsmodellanalysen notwendig sind. Desweiteren wird untersucht, inwieweit mit Diffusionsmodellen längerdauernde Entscheidungsprozesse analysiert werden können. Die Ergebnisse sollen dabei helfen, das Anwendungsgebiet dieser Form der Datenauswertung von den typischen Reaktionszeitaufgaben der experimentellen Psychologie auf das Feld der Entscheidungsforschung zu erweitern.
Self-Regulatory Processes in Danger Perception and Risky Decision Making (2013-2014)
The fast detection of signals for impending dangers and the avoidance of risky situations are of special importance for human information processing because these strategies might help preventing possible negative events, losses, injuries, or death. In the last decades numerous psychological studies provided evidence for a fast and efficient system of danger detection, which promotes an automatic allocation of attention on possible danger cues (e.g., Öhman, Lundqvist, && Esteves, 2001; Pratto && John, 1991). Many authors argue that such a system might be a product of evolution because it helped in the past to ensure survival. However, this argument is somewhat flawed because also the rapid detection of positive signals can bring survival benefits; for example, the immediate detection of a prey animal or - in the case of danger - of a hiding place or an escape route can be crucial to survive. This position is supported by recent findings that demonstrate positive biases in perception and decision making (e.g., Balcetis && Dunning, 2006; Becker, Anderson, Mortensen, Neufeld, && Neel, 2011).
On a broader perspective, the assumption of a general stable bias that promotes the allocation of attention either always on negative or always on positive stimuli seems to be too inflexible and thus of limited adaptive value. A flexible system, however, directing the attention to stimuli that are specifically relevant in a given situation will always be superior. In line with this argumentation, we assume that self-regulatory processes promote such a situation-specific adaption of attentional processes. The aim of the present project is to analyze how the perception of danger signals and risky decision making depend on characteristics of the present situation.
One important variable that is analyzed in this context is the amount of control over potential dangers. We argue that it is only important to focus on danger cues if the perceived information can help to prevent any negative consequences. The early perception of signals for uncontrollable dangers -on the contrary - would be maladaptive because cognitive resources are wasted, anxiety or other negative emotions are triggered and psychological wellbeing is reduced, without helping to avoid the impending dangers. Therefore we expect that perceived control triggers self-regulatory processes that in turn direct attention to danger cues. The lack of control and the experience of helplessness should reduce sensitivity for danger signals.
A second process that is addressed with the current project regards the relevance of outcomes for the actor. We expect that negative outcomes are perceived to be more likely when this risk applies to the own person, and less likely, when other persons are affected. Thus, if risks are relevant for the own person, risk aversion is predicted, while more risk seeking behavior might be found whenever others money is at stake.
In the present project, the influence of perceived control and of self-relevance of decisions on the perception of danger signals and on risky decision making is analyzed with four behavioral experiments using visual search tasks, eye-movement recordings, and behavior-economical paradigms.